Research On ZYNQ-based Relative Pose Measurement Of Space Non-cooperative Targets And Its Onboard System

This paper focuses on the relative position attitude estimation problem of non-cooperative target satellites based on a single optical sensor during the recovery of failed satellites in orbit by space active pursuers.Considering the limitations in power consumption and stability of edge-end controllers in space,this paper focuses on ZYNQ-based bare-metal monocular vision space non-cooperative target relative position attitude estimation key technology.The main research contents include:(1)The system hardware architecture is designed with Xilinx ZYNQ-7000 SoC as the core.The monocular vision image processing algorithm system and deployment scheme based on OV5640 visible light camera are developed.The system processing speed comparison results show that the ZYNQ-based hardware and software co-design accelerates significantly,giving full play to the advantages of the FPGA high energy efficiency ratio.(2)Based on the self-built training data set and test data set,the HOG-SVM target recognition algorithm and the testing of the algorithm on ZYNQ side are studied,based on which the algorithm processing efficiency and stability are improved,and the target satellite recognition accuracy and recall rate are increased to 94.1%and 97.4%respectively compared with the traditional target recognition algorithm.(3)A feature extraction method based on Hough linear transformation is proposed.The results show that the position error rate is controlled within 4%and the error of Euler angle is controlled within about 2°.The results show that the position error rate is within 4%and the Euler angle error is about 2°.The efficiency of feature extraction based on the Hough linear transform is greatly improved,and the processing time is only 0.76ms per frame.(4)The Kalman filter-based target pose and motion state estimation algorithm is implemented to address the drift and noise of pose results,and the error rate of rotation angular velocity estimation is 3.283%,and it plays a feedback role for pose solution.Through the above research and discussion,the results show that the above target identification algorithm has strong robustness and accuracy,and the positional estimation scheme fully meets the requirements of system operation efficiency.